[BUG] Object Detection AutoTrain Error: iteration over a 0-d tensor

[rileybolen]

Prerequisites

• [X] I have read the documentation.
• [X] I have checked other issues for similar problems.

Backend

Hugging Face Space/Endpoints

Interface Used

UI

CLI Command

No response

UI Screenshots & Parameters

Error Logs

100%|██████████| 13/13 [00:10<00:00, 1.51it/s]/app/env/lib/python3.10/site-packages/autotrain/trainers/object_detection/utils.py:158: UserWarning: Creating a tensor from a list of numpy.ndarrays is extremely slow. Please consider converting the list to a single numpy.ndarray with numpy.array() before converting to a tensor. (Triggered internally at /opt/conda/conda-bld/pytorch_1712608935911/work/torch/csrc/utils/tensor_new.cpp:274.) batch_image_sizes = torch.tensor([x["orig_size"] for x in batch]) INFO: 10.16.9.183:64413 - "GET /ui/accelerators HTTP/1.1" 200 OK INFO: 10.16.27.38:51108 - "GET /ui/is_model_training HTTP/1.1" 200 OK ERROR | 2024-05-24 14:09:32 | autotrain.trainers.common:wrapper:120 - train has failed due to an exception: Traceback (most recent call last): File "/app/env/lib/python3.10/site-packages/autotrain/trainers/common.py", line 117, in wrapper return func(*args, *kwargs) File "/app/env/lib/python3.10/site-packages/autotrain/trainers/object_detection/main.py", line 199, in train trainer.train() File "/app/env/lib/python3.10/site-packages/transformers/trainer.py", line 1885, in train return inner_training_loop( File "/app/env/lib/python3.10/site-packages/transformers/trainer.py", line 2311, in _inner_training_loop self._maybe_log_save_evaluate(tr_loss, grad_norm, model, trial, epoch, ignore_keys_for_eval) File "/app/env/lib/python3.10/site-packages/transformers/trainer.py", line 2721, in _maybe_log_save_evaluate metrics = self.evaluate(ignore_keys=ignore_keys_for_eval) File "/app/env/lib/python3.10/site-packages/transformers/trainer.py", line 3572, in evaluate output = eval_loop( File "/app/env/lib/python3.10/site-packages/transformers/trainer.py", line 3854, in evaluation_loop metrics = self.compute_metrics(EvalPrediction(predictions=all_preds, label_ids=all_labels)) File "/app/env/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context return func(args, **kwargs) File "/app/env/lib/python3.10/site-packages/autotrain/trainers/object_detection/utils.py", line 188, in object_detection_metrics for class_id, class_map, class_mar in zip(classes, map_per_class, mar_100_per_class): File "/app/env/lib/python3.10/site-packages/torch/_tensor.py", line 1047, in iter raise TypeError("iteration over a 0-d tensor") TypeError: iteration over a 0-d tensor

ERROR | 2024-05-24 14:09:32 | autotrain.trainers.common:wrapper:121 - iteration over a 0-d tensor INFO | 2024-05-24 14:09:32 | autotrain.trainers.common:pause_space:77 - Pausing space...

33%|███▎ | 100/300 [01:32<03:04, 1.08it/s]

Additional Information

The training is able to start and make some progress, but it seems that after the first epoch of training is completed the training fails with this error.

[abhishekkrthakur]

did you also upload validation data or just training data?

[rileybolen]

I only uploaded training data, it looked like it automatically did the train/val split. I did find an image that was listed in my metadata twice, so I am wondering if maybe it one of those entries ended up in validation and one in training, causing the image to not be found in the validation set. I fixed this and I am trying again. I can also try manually splitting and uploading my validation data. I will let you know if that fixes the error.

[abhishekkrthakur]

it does auto splitting. that shouldnt be an issue.

I did find an image that was listed in my metadata twice, so I am wondering if maybe it one of those entries ended up in validation and one in training, causing the image to not be found in the validation set. I fixed this and I am trying again

please let me know. this case should be caught earlier

[rileybolen]

@abhishekkrthakur I tried removing the duplicated image record from metadata.jsonl and I still got the same error.

[abhishekkrthakur]

okay. so the issue is happening for datasets that have a single class. im fixing the issue and will update here asap. i really hope it works for you end to end now. and deep apologies.

[rileybolen]

@abhishekkrthakur Sounds good, thanks! And no problem, I'm glad I can help test a new feature.

[abhishekkrthakur]

just pushed a fix and tried on my own please make sure you are on v0.7.110 or above.

[abhishekkrthakur]

please let me know if you still face issues

[rileybolen]

It seems that the training has worked, thanks! I am just facing issues now with the Serverless Inference API, but I think that is separate from this repo. So I think this issue is solved now!

[abhishekkrthakur]

The api wont work immediately. Try a few minutes after training is done :) and thank you so much for all the help :)

[abhishekkrthakur]

@rileybolen thank you very much for helping debugging this and apologies for the inconveniences. As a gratitude, we have added a $25 credit to your hugging face account that you can use for spaces, inference endpoints, autotrain or other huggingface services.

[abhishekkrthakur]

fixed